Maybe three little mistakes in file 'particle_system.py'

[Rabmelon]

I am learning this program to finish my taichiCourse01 final project. And I found three confused lines in file 'particle_system.py':

1. in line 179, I think it should be 'particle_diameter' but not 'self.particle_radius'. I guess this also leads to an overlap while drawing, so you used 'radius=ps.particle_radius / 1.5 ...' in 'demo.py'.
2. in line 178, I think there should be a tiny number added after 'lower_corner[i] + cube_size[i]', so you won't lose the particles in the top row and the right colume.
3. in line 120, I think it should be 'continue' but not 'break', because a break there will lead to a lack of neighbour search. But after I change it to continue, the para 'offset' won't update. As to say, for a para 'center_cell' = [0, 1], para 'cell' should be [-1, 0] then [-1, 1] then ... But actually, para 'cell' will loop as [-1, 0] and won't update to [-1, 1], so this caused an endless loop. Besides, I created a code as follows for test and it went well. I am confused what's the problem and will really appreciate your solution.

import taichi as ti
ti.init(arch=ti.cpu)

dim = 2

center_cell = ti.Vector.field(2, dtype=ti.i32, shape=2)
center_cell[0] = ti.Vector([0, 1])
center_cell[1] = ti.Vector([2, 2])

@ti.func
def is_valid_cell(cell):
    flag = True
    for d in ti.static(range(dim)):
        flag = flag and (0 <= cell[d] < 3)
    return flag

@ti.kernel
def func_test():
    for i in range(2):
        print(center_cell[i], ';')
        for offset in ti.grouped(ti.ndrange(*((-1, 2),) * dim)):
            cell = center_cell[i] + offset
            print(cell, end=', ')
            if not is_valid_cell(cell):
                print('kill', end='! ')
                continue
        print()

func_test()

with an output: [0, 1] ; [-1, 0], kill! [-1, 1], kill! [-1, 2], kill! [0, 0], [0, 1], [0, 2], [1, 0], [1, 1], [1, 2], [2, 2] ; [1, 1], [1, 2], [1, 3], kill! [2, 1], [2, 2], [2, 3], kill! [3, 1], kill! [3, 2], kill! [3, 3], kill!

environment: [Taichi] version 0.8.1, llvm 10.0.0, commit cc2dd342, win, python 3.8.1

[erizmr]

Thanks for reporting this. I will take a look and come back to you.

[erizmr]

Hi, I have check the problems you mentioned.

• Problem 1, I think it is ok to change the particle_radius to particle_diameter. The overlapping is actually not a mistake since the particles here are sampling points of kernel functions.
• Problem 2, good point.
• Problem 3, good catch, I agree that it should be continue but not 'break' but I cannot reproduce the endless loop issue you mentioned after I changed it to continue, Would you mind providing the code that can reproduce the "endless loop"? Thanks.

[Rabmelon]

Thank you for the reply.

• For problem 1, thank you for the remind.
• For problem 3, sorry for my fault that I forget my earlier modification of gridsize = 2 * support radius and my misunderstanding of the para padding before. The endless loop will occur when the initialized particles invade the padding region. Anyway, particles will not come into the padding region after correctly initialized, so the problem won't occur. code_for_problem_3.zip

Again thank you for your time and help!